Mwi Pref

The many-worlds interpretation of quantum mechanics: psychologicalversus physical bases for the multiplicity of worlds.

Howard Barnum

Los Alamos National Laboratory, MS B256, Los Alamos, NM 87545

Preface

This hitherto unpublished 1990 paper presents my views on the many-worlds, or, as I prefer tocall it, following Everett, the relative state interpretation of quantum mechanics. It was rejectedby one journal in 1990, primarily on the (I think incorrect) grounds that it was too similar to theAlbert and Loewer many minds version. It has seen limited circulation since then. My viewsare quite similar to ones since put forth by Simon Saunders, and subscribed to by David Wallace,among others. I believe that, within the relative state interpretation, when an observer interactswith a system in a measurement process in such a way that the combination of the two evolvesinto a superposition of the observer seeing macroscopically different apparatus pointer readings(or other macroscopic states), the observers consciousness forks or splits, because only thosesubspaces of Hilbert space describing definite macroscopic measurement results satisfy psycho-logical/philosophical conditions for supporting a unified consciousness persisting through time.These conditions probably include things like the persistence of memory, and are probably closelylinked to the decoherence of certain subspace decompositions or approximate decompositions,corresponding to persisting macroscopic classical structures that can support such persistingmemoriesrelating the approach to the decoherence-based ones of, for example, Zurek, and Zeh,though this relation was not much stressed in the paper.

I still agree with most of what is in the paper, with some notable exceptions. First, the dis-cussion of how one might have empirical evidence for nonprobabilistic indeterminacy seemsshaky: the evidence cited might just as well be taken to be evidence against independent trials,or against exchangeability. (This does not vitiate the point that relative state-ers may need toestablish not just that if there are probabilities they must be the Born ones, but that (on the relativestate interpretation) the indeterminism facing a quantum experimenter about what will be subjec-tively experienced as an outcome is correctly viewed as probabilistic at all.) Second, as I think the1990 referee pointed out, the improvement on Everetts derivation of probabilities (sec. 8.2),while it uses Gleasons theorem to avoid Everetts assumption that the probability of a wavefunc-tion component can depend only on its squared modulus, does not justify the noncontextualityassumption implicit in Gleasons theorem. There are also more minor exceptionsfor instance,I now see somewhat less value in de Broglie-Bohm-Bell nonlocal hidden variable theories thansuggested toward the end of this paper.

Note that while I think it has much to teach us about quantum mechanics, I am not a committedbeliever in the relative state interpretation.

The following pages were scanned from a hard copy, and the address in them is obsolete.

Los Alamos, December 2002 and January 2004

[email protected]

THE MANY-WORLDS INTERPRETATION OF QUANTUM MECHANICS:

PSYCHOLOGICAL VERSUS PHYSICAL BASES FOR THE

MULTIPLICITY OF "WORLDS".

Howard N. Barnum 6401 Academy NE, #168 Albuquerque, NM 87 109 (505) 828-2567

THE MANY-WORLDS INTERPRETATION OF QUANTUM MECHANICS: PSYCHOLOGICAL VERSUS PHYSICAL BASES FOR THE

MULTIPLICITY OF "WORLDS".

1

Abstract. A crucial distinction in discussions of the many-worlds interpretation of quantum

mechanics (MWI) is that between versions of the interpretation positing a physical multiplicity

of worlds, and those in which the multiplicity is merely psychological, and due to the splitting

of consciousness upon interaction with amplified quantum superpositions. It is argued that

Everetts original version of the MWI belongs to the latter class, and that most of the

criticisms leveled against the MWI, in particular that it is illogical or incoherent, are not valid

against such psychological multiplicity versions. Attempts to derive the quantum-mechanical

probabilities from the many-worlds interpretation are reviewed, and Everetts initial derivation

is extended to show that these are the unique possible probabilities. But there remains a

challenge for proponents of the MWI: to show that their interpretation requires probabilities,

rather than merely nonprobabilistic indeterminacy.

1

Introduction. While it has enjoyed some popularity, particularly among cosmologists, the

relative state or many-worlds interpretation of quantum mechanics (henceforth MWI) appears

to have fallen into disfavor among those concerned with the problem of interpreting quantum

theory. D'Espagnat (1976) has criticized the theory in his seminal book, and Wheeler (1977),

one of the best known proponents, has renounced it. Bell has said that "...the many-universes

interpretation is a kind of heuristic, simplified theory, which people have done on the backs

of envelopes but haven't really thought through. When you do try to think it through it is

not coherent." (Bell, interviewed in Davies and Brown (1986)), and his view appears fairly

widespread. Even among those who view the MWI as consistent, virtually no-one anymore

believes with Everett that it shows that "the Schrodinger equation formalism yields its own

interpretation." The major objective of this paper is to show that despite such criticism, the

MWI is indeed a coherent and consistent interpretation of quantum mechanics arising naturally

from the Schrodinger formalism, by developing a "psychological" version of the MWI in

which the multiplicity of "worlds" appears as a multiplicity of perspectives on a single

physical reality as seen from the points of view of different conscious subsystems' of this

physical reality. This physical reality will be described by the usual quantum mechanical

apparatus of a Hilbert-space statevector evolving over time according to the Schrodinger

equation, with no extra physical structure introduced to account for "branching universes," and

will be shown to give rise to the multiplicity of "worlds." It will also be shown to imply that

there is only one acceptable probability distribution over them, which coincides with the

quantum mechanical one.

The term 'subsystem' is here meant in its general sense, not necessarily in its special quantum mechanical sense of one factor of a tensor product of Hilbert spaces.

1

2

Sections 1 and 2 provide preliminaries, with section 1 reviewing the reasons for

wanting an interpretation, such as the MWI, based solely on the Schrodinger equation and

rejecting the projection postulate, and section 2 setting forth a variety of such interpretations,

including different possible "many-worlds" interpretations, for comparison with the

"psychological" version of the MWI which is the main subject of the paper. Section 3

develops this psychological version in detail, and argues that it coincides with Everett's views,

while sections 4 and 5 develop it further by comparing its approach to psychological unity through time and across subspaces of Hilbert space to that of other versions of the MWI--

in the process setting to rest the fear, expressed by Bell (1981), that the MWI indulges in

"solipsism of the present moment." Section 6 reviews criticisms of the MWI and attempts to

answer them in the light of the psychological version. Sections 7 through 9 take up the

question of the relationship between the MWI and the probabilities defined by the Born

interpretation of the wavefunction. The key points in the discussion are section 8's argument

that the Born probabilities constitute the only permissible probability distribution over different

"worlds" corresponding to different observed experimental outcomes, and section 9's argument

that we cannot yet rule out the possibility of nonprobabilistic indeterminacy. A variety of

considerations bearing on the choice between the MWI and other "no-statevector-reduction"

interpretations are noted as they arise throughout the paper, and some are touched on again

in the concluding section. But the main thrust of the paper is that whatever such

considerations may suggest about the choice between the MWI and other interpretations, the

only problem with the consistency or coherence of the MWI lies in the need to arbitrarily

postulate that the indeterminacy of which branch of the statevector one will end up in is of

a probabilistic, rather than nonprobabilistic nature. The lack of an argument for this postulate

based in the MWI itself, remains an obstacle to the claim that the MWI provides a consistent,

3

coherent interpretation of quantum mechanics which can generate the quantum mechanical

predictions from the Schrodinger equation formalism without any extra structure or postulates.

1. Reduction, Schrodinger Evolution, and the Many-Worlds Interpretation. The MWI can

be viewed as an attempt to answer the question: why don't we observe superpositions of

macroscopically different states? Schrodinger pointed out, with his cat example, that the

Schrodinger equation implied that such states could be prepared. The "orthodox"

interpretation's answer to the question of why we don't observe them was to postulate a mode

of time evolution distinct from the continuous, unitary linear evolution according to the

Schrodinger equation, that applied whenever a measurement was made on a system, throwing

it from a superposition into one of the eigenvectors of the measured observable (with

probabilities given by the squares of the amplitudes of the components of the statevector

expanded in terms of these eigenvectors). This "collapse" or "reduction of the wavepacket"

creates several difficulties. Particularly bothersome are: (1) there is no rigorous, physical

specification of the conditions under which collapse, rather than Schrodinger evolution, occurs,

and (2) gedankenexperiments in which one tests whether collapse has occurred by looking for

the interference between the superposed alternatives which, on a view rejecting reduction,

would persist after measurement, suggest such interference would be observed.2 In (2), I am

thinking specifically of the reversal of the Stern-Gerlach experiment. (For classic discussions

of this, see Wigner (1967) and Bohm (1951); for a modern treatment, see Englert, Scully, and

Schwinger (1988) and even permutations thereof (1988), (1989)J While it has been argued

Other difficulties are: (3) the orthodox view seems to require the postulation of a non-quantum-mechanically described outside system "observing" a quantum mechanical system, creating difficulties describing the whole universe in quantum- mechanical terms, and (4) it is probably not possible to describe the "reduction of a statevector" in a Lorentz-invariant way (see Aharonov and Albert (1980, 1981)).

2

4

that thermodynamic considerations make even the partial reversal of macroscopic measurements

impracticable (Daneri, Loinger, and Prosperi (1962); Peres (1980)) that does not resolve the

issue of principle.

What inclines us to accept reduction, rather than continual Schrodinger evolution, is that

we dont directly observe macroscopic superpositions. One might make an acceptable theory

by making physically precise the conditions under which reduction occurs, either by describing

an instant where it occurs, or by slightly modifying the Schrijdinger equation so that it remains

as it is for microscopic phenomena unobserved by macroscopic apparatus, but shades

(continuously?) into statevector reduction as the system interacts with larger and larger apparati

capable of correlating its state with more and more macroscopic phenomena. Either of these

possibilities are in principle, experimentally distinguishable from pure Schrodinger evolution.

But if we can find an interpretation of the quantum mechanical apparatus of state-vector and

Schrodinger equation which implies that we wont observe macroscopic superpositions, then

we wont have to assume reduction. Both Bohms (1952) theory (which adds a bit to the

quantum mechanical apparatus, but doesnt alter its experimental implications) and the MWI

are attempts to do this, in different ways.

2. Different Versions of the MWI and Related Interpretations. Different proponents of

the MWI present different versions, and are not always clear about what they have in mind.

But clarity about ones interpretation and the assumptions behind it are essential to an adequate

assessment of it. I will therefore outline several possible versions of the MWI, and some

related interpretations which will help clarify the MWI by contrast and which I regard as its

closest competitors. All of them attempt to solve the measurement problem by rejecting

statevector reduction. Some of these versions were suggested by the incisive analysis of Bell

5

(1981) but I will concentrate on a distinction which Bell neglected, that between "physical"

and "psychological" interpretations of the multiplicity of worlds.

(A) The first version of the MWI is the simplest, and perhaps the closest to Everett's. The

structure of the physical world corresponds just to the statevector in Hilbert space, evolving

according to the Schrodinger equation. The multiplicity of "worlds" is really a multiplicity

of consciousnesses, not a multiplicity of physical systems, and the "branching" that occurs is

also an essentially psychological process that occurs when a consciousness observes a

superposed system: the conscious system goes into superposition, as the Schrodinger equation

implies, but thereby splits into several consciousnesses. There is no multiplication of physical

worlds, only a complexification of the structure of the statevector of the universe, which has

differentiated so as to support, indeed given the psychological conditions for the unity of

consciousness to require, several consciousnesses where before there was only one. This

interpretation will be my primary focus, and I will expound it at length below. (It might be

clearer to call this the Relative State Interpretation, reserving the term 'Many-Worlds' for

physical-multiplicity interpretations like B and C below; I refer to both types of interpretations

as versions of the MWI, since this has become the established practice.)

(B) Realism about the statevector as above, plus the statement that there is a physical

multiplicity of worlds. The square modulus of the projection of the statevector on an

eigenvector of an observable Z is stipulated to give us the proportion (or relative measure) of

worlds in which a measurement of that observable yields the corresponding eigenvalue. (The

usual probability assertions of quantum mechanics may then be derived from the assumption

6

that we are equally likely to be in any one of the existing worlds). Deutsch (interviewed in

Davies and Brown (1986)) may have in mind something of this sort.

(C) Interpretation B, with the stipulation that the worlds that exist correspond to position

eigenvectors; this is essentially Bell's (198 1) version of Everett's (1957a,b) MWI. This gets

rid of the difficulty with (B), that it introduces "observation" into the fundamental physics

again in the choice of an observable whose eigenvectors determine the nature of the physically

different worlds.

(D) Interpretation (B), plus a linking of particular worlds at one time with worlds at other

times, into physical trajectories. Just how to select which worlds get linked into trajectories

is a problem seemingly susceptible to many solutions; one desirable criterion might be an

"anti-solipsism" requirement: that what is in our memory of "events at time t-1" at time t,

correspond to what happened in this branch at time t-1. (It would seem there is no way we

could observationally distinguish theories which violate this requirement from those that don't,

however.) A great variety of specifications of trajectories may be possible; a particularly

interesting specification gives rise to trajectories like those in the de Broglie-Bohm pilot wave

model.

(E) B, but with the various worlds being merely possible worlds, and the real world being

chosen with a distribution llv(z)I12 out of the class of possible worlds, where z ranges over

the eigenvalues of the relevant observable Z. This is another of Bell's suggestions, though

it is of course no longer a many-worlds interpretation.

7

(F) Like (E), but at each moment the actual world is chosen out of the comparison class of

possible worlds with distribution lIw(x) 112, where x ranges over the eigenvalues of position.

(G) Like (D), but there is a single actual trajectory. For a particular choice of form for the

trajectories, interpretation G becomes the de Broglie-Bohm pilot wave theory of Bohm (1952).

E,F, and G are of course not many-worlds theories (I would call them all hidden-

variable theories); A through D are.

Do any of these interpretations correspond to the classic expositions of the MWI by

Everett, DeWitt and others (De Witt, and Graham 1973)? Despite formalized expositions like

Everett's "relative state" formulation, the MWI is commonly thought to be rather unclear. The

root of the unclarity is Everett's claim that the MWI is the natural interpretation of the

quantum apparatus of statevector and Schrodinger equation, to which it adds nothing. This

has seemed to conflict with talk of "branching universes" which appear to be an added

structure imposed on the evolving statevector, since after all components of statevectors are

not linked through time in the quantum formalism in any kind of tree structure. Moreover,

if the additional structure really is a branching tree, the problem of reduction of the

wavepacket has not really been avoided, but merely reformulated: the splitting of a universe

appears to be basically the same process as reduction, with all the same problems as to when,

precisely, it occurs. These difficulties stem from a fundamental ambivalence among critics and

some proponents of the MWI about whether the "many worlds" reflect a physical multiplicity

of universes, or simply a division of consciousnesses within one physical system. I will

argue that Everett explicitly endorsed the latter view.

8

3. Version (A) of the MWI: Psychological Multiplicity Without Physical Multiplicity.

3.1 Psychophysical parallelism. The MWI as presented by Everett is an attempt to interpret

the statevector and Schrodinger equation without any extra mathematical formalism (and in

particular, without the either the projection postulate or a physical multiplicity of worlds). I

read Everett as endorsing version (A) of the MWI. On this reading there is one universe,

represented by a statevector, which may however be a superposition of infinitely many

macroscopically or microscopically different states. Evolution of this state vector is always

according to the Schrodinger equation. To the question "Why don't we observe macroscopic

superpositions like Schrodinger's cat?" a simple reply may be given. One system measures

(observes, if it is conscious) another when its state becomes correlated with the other's through

Schrodinger evolution. That is, letting y' represent the apparatus (or observer) state and $

represent the observed system state, when the state of the total system A + S evolves from ~ o @ ( ~ l ~ ' , ) to ~,(tfI@u/s,). (v", represents the initial, nonregistering state of the apparatus, while the VI are distinct states which may be thought of as pointer positions (or states in which the observer sees a definite pointer position) or the like.) In attempting to observe a

macroscopic superposition like Schrodinger's cat, my state would become correlated with the

cat's, and I would go into a superposition of states "me seeing a dead cat" and "me seeing

a live cat." Such a macroscopic superposition does not possess the necessary unity to be a

"state of consciousness" (using 'state' in a non-QM sense here). Therefore what happens is

my consciousness splits into two consciousnesses, for the components "me seeing a dead cat"

and "me seeing a live cat," which are both elements of physical reality, do possess the unity

required to be states of consciousness. And these components (and their subsequent histories,

until their next encounters with amplified quantum phenomena) moreover possess the necessary

continuity (of memory-contents, e.g.) with earlier and later states to form the history of a

9

consciousness persisting through time-- with the one bizarre aspect that when I observe an

amplified quantum superposition (when I interact with it via a Hamiltonian which correlates

components of my state to components of the superposition) my consciousness splits or forks

into many branches. (Theoretically, though as a practical matter it would be impossible,

someone could perform on me the analog of the reversal of the Stern-Gerlach experiment, and

the branches could join again, but in such a process the results of measurement would be

erased from my memory(s)--along, perhaps, with everything else that happened between the

splitting and rejoining.) These branches are branches of my consciousness, and the branching

is not a physical process, but one that occurs because of the conditions for personal identity,

or identity of consciousness, which are more matters of philosophy or psychology than physics.

It may seem strange to invoke psychological or philosophical criteria in explaining the

results of physical processes, but it is not really. Any physical theory needs some crude

account of how the theory links up with our observations, in order to be of interest to US

(though of course the theory's meaning is not exhausted by statements about observations).

These include crude theories about our eyes, bodies, sense-organs, and how they perceive what

they do. Advocates of the MWI could argue that in order to get theory properly related to

observation (part of interpreting the theory or mathematical formalism) in quantum mechanics,

we simply need a slightly more sophisticated "stylized psychology" which takes into account

conditions for the unity of conscious experience. This does not mix psychology into physical

theory, so that we need psychological concepts like mind in the statement of the basic physical

theory; we simply need to know a bit about the way psychological states supewene on

physical states in order to get the observational implications of the basic physical theory

(which however contains no psychological concepts) right. By contrast, it might be argued

10

that some proponents of versions of the more "orthodox" interpretation of QM, with the

reduction postulate, do invoke psychological concepts at the level of basic physical theory:

for example Wigner (1967, 1976), who suggests (in part because of his acceptance of the

reverse S tern-Gerlach argument against early reduction of the statevector) that reduction must

not be a purely physical process, but must rather be an effect of nonphysical mind on matter.

This sort of thing is worrisome in physical theory (particularly as we have no precise way of

deciding when we are in the presence of a "mind" or Cartesian "soul"), but the MWI is not

guilty of it.

3.2 Everett as a Subscriber to Version (A)

Everett spends a lot of time on formalized relative state theory, and relatively little on

the assumptions about psychophysical parallelism which I have argued are crucial to the

success of the MWI. It is these assumptions which give certain "relative states" (states

relative to particular states of conscious observers) their interest as potential "worlds" from the

point of view of particular consciousnesses. Among supporters of the MWI, Zeh is most

explicit about this: "The different components represent two completely decoupled worlds.. .. As the "other" component cannot be observed any more, it serves only to save the consistency

of quantum theory .... This interpretation, corresponding to a "localization of consciousness" not only in space and time, but also in certain Hilbert-space components, has been suggested

by Everett ..." (Zeh 1970) Everett himself, while stressing the importance of psychophysical parallelism and of representing the observer as a physical system within the theory, does not

explicitly discuss the role of conditions for the unity of consciousness. There are strong hints

that the view sketched above is implicit in his thought, however:

11

We thus arrive at the following picture: Throughout all of a sequence of observation processes there is only one physical system representing the observer, yet there is no single unique state of the observer (which follows from the representations of interacting systems). Nevertheless, there is a representation in terms of a superposition, each element of which contains a definite observer state and a corresponding system state. Thus with each succeeding observation (or interaction), the observer state "branches" into a number of different states. Each branch represents a different outcome of the measurement and the corresponding eigenstate for the object-system state. All branches exist simultaneously in the superposition after any given sequence of observations. (Everett 1957a)

In a footnote, he adds "This total lack of effect of one branch on another also implies

that no observer will ever be aware of any "splitting" process.

This passage is somewhat ambiguous, and I suspect different interpretations of it and

passages like it (Everett 1957b, p. 62n.) have given rise to many different suppositions about

what Everett's MWI really is. One thing that suggests the splitting is a psychological splitting

of consciousness, rather than some sort of physical splitting of one world into many, is that

Everett clearly says that "the observer state "branches"" rather than, say, "the universe

"branches"." That Everett always puts the terms 'split' and 'branch' in quotes suggests the

same thing. Also, Everett states that throughout the process, there is only one physical system

representing the observer.

4. Psychological Unity Through Time in Various Versions of the MWI. Bell's

strongest objection to the MWI is that he believes it implies a "solipsism of the present

moment," in that the MWI shows that for an observer in given branch of the statevector at

a given time, all of his memory-contents will be (statistically speaking) consistent with the

usual quantum-mechanical predictions, but, Bell believes, there is no linking up of these time-

slices of consciousnesses into histories: no guarantee that the "I" that has certain experiment-

results recorded in its memory actually had those experiences (Bell 1981). Although Everett's

12

scheme implies the existence of an "I" at a previous time who had those experiences, Bell

thinks there is no physical identification of this "I" with the current "I" who remembers the

experiences. I think this problem stems from a misunderstanding of the relative importance

of physics and psychology/philosophy in the MWI. The primary aim of the MWI was

precisely to show that an interpretation of QM observationally indistinguishable from the

orthodox one could be constructed by adding to the Schrodinger equation formalism only some

assumptions about psychological or philosophical criteria for identity of consciousness. A

minimal way of attempting this gives rise to Bell's time-slices. But if we have qualms about

the metaphysical sparseness of the world thus construed, we may add additional suppositions

about criteria for identity of consciousness through time, and recover a world of persisting

(though branching) individual consciousnesses whose current memory-contents are generally

reliable indicators of past experiences. (Moreover, we can get this unification through time

whether we have a type (A), (B), or ( C ) version of the MWI: that is, whether one supposes

the multiplicity of worlds at any given time is physical or psychological.) Thus if we want

to require (1) some degree of physical causal connectedness and (2) some degree of

consistency of content, such that memories generally correspond to earlier experiences, as

conditions for identity of consciousness through time, we can link up the "right" components

of the statevector into consciousnesses with branching histories, which are not solipsistic in

the way Bell was worried about. Each component of the universal statevector at time t is

causally linked to components at later times (requirement l), and some of these later

components share enough conscious content (requirement 2) that they may be linked into a

conscious history. It is a mistake to suppose that physics should be expected to supply the

extra structure which ensures these trajectories link up. (Though if the physics does add

13

trajectories, as in (D), we will want them to be the right ones for psychological continuity.)

At one point Bell likens Bohm's theory to the MWI plus trajectories (and of course,

the singling out of one trajectory as the physically real one). He sees a major advantage for

Bohm's theory in the specification, in the physics, of a trajectory--a single history for the

world, which provides a basis for rejecting solipsism. Presumably the idea is that now I know

I am identical with the previous person whom physics says existed and had experiences I now

remember having, because there aren't all these other "versions of me" in other universes who

I might worry that I'm really, metaphysically continuous with rather than the one whose

experiences I remember. But this problem isn't unique to the MWI (There are other people

even in a single branch of this universe and how, after all, do I know I wasn't really,

metaphysically one of them a few seconds ago, instead of the person whose experiences I

remember? How do I know I was really, metaphysically ANYBODY a few seconds ago, even

if there are no other "versions of me in other branches of the universe" to worry about?). I

think Bell is here expecting physics to solve puzzles which it need not solve. Finally, it is

important that we can introduce psychological "trajectories" in version (A), since as I will

show in Section 8, this is useful in the derivation of the correct probabilities from version (A),

a crucial part of any argument for the observational indistinguishability of version (A) from

standard quantum mechanics.

5. Psychological Unity and Multiplicity Across Subspaces of Hilbert Space in Various

Versions of the MWI. The considerations of the previous sections suggest two ways of

viewing the measurement process, corresponding to different views of the psychophysical

relationship between brains in different Hilbert-space components and consciousnesses. In one,

14

there is a single consciousness before the interaction, splitting into several afterwards. In the

other, there are the same number of consciousnesses before and after the interaction: one for

each component of the Hilbert space decomposition in terms of eigenvectors of the relevant

observable. Everett appears to plump for the former, and it indeed version (A) of the MWI

appears to provide no basis for a psychological multiplicity of consciousnesses before

measurement. In the case of type (B) or (C) theories, which have physical mutiplicity but no

trajectories, one might imagine either that consciousness unifies across physically distinct

worlds in which its state is identical, or that it remains numerically distinct.

6. Critics of the MWI.

6.1 Bell. Bell's views on Everett, as I have suggested above, provide an example of how an

excessively physical interpretation of Everett's statements about "branching" and "multiple

worlds" can lead to trouble. Bell's view of Everett's theory is summarized in the statement

that:

The Everett (?) theory of this section will simply be the pilot-wave theory without trajectories. Thus instantaneous classical configurations x are supposed to exist, and to be distributed in the comparison class of possible worlds with probability lh@ll. But no pairing of configurations at different times, as would be effected by the existence of trajectories, is supposed. (Bell 198 1)

Bell then makes a series of points, labelled (A) through (D), about how this Everett (?)

theory seems to differ from what Everett (and/or DeWitt) actually say. I think it is fair to

say that Bell supposes a coherent version of the Everett theory must differ from what Everett

and DeWitt seem to 'be saying on these four points. But in my view, Bell misinterprets

Everett, at least, on these points, precisely because he takes his statements too physically.

Bell's point (A) is that it is unnecessary to multiply worlds. Bell begins by making

what I have taken to be the main point of the MWI, that "The psychophysical parallelism

15

is supposed such that our representatives in a given "branch" universe are aware only of what

is going on in that branch." But he goes on: "Now it seems to me that this multiplication

of universes is extravagant, and serves no real purpose in the theory, and can simply be

dropped without repercussions. So I see no reason to insist on this particular difference

between the Everett theory and the pilot-wave theory-- where, although the wave is never

reduced, only one set of values of the variables x is realized at any instant." Now this is

an interesting point--that one could throw out all worlds but one, and if it were true that one

would lose nothing essential, one might want to do it. Everett and DeWitt believe that one

does lose something; the existence of all branches is essential to the Everett-DeWitt-Graham

demonstrations, discussed by Bell under (C), that one can recover the Born statistical

interpretation from the formalism itself. Since Bell thinks this demonstration is unsuccessful,

I assume he therefore isn't worried about losing this. From his point of view, both the

hidden variables interpretations and the MWI have to make an additional assumption

introducing the correct probability measure. Bell's Everett(?) (and Bohm) do indeed add

such an assumption about the selection of the actual world from an ensemble of possible

worlds with probability measure on that ensemble given by Ilv211. Moreover for Bell, who sees both the MWI and de Broglie-Bohm as introducing a structure of "worlds" or

trajectories in addition to the statevector, we can remove the other worlds from this extra

structure without losing the possibilities of interference present in the statevector; for Everett,

the statevector is all there is, and we cannot get rid of these other worlds without destroying

these interference possibilities.

Another problem is discussed by Bell under (B):

it could be said that the classical variables x do not appear in Everett and DeWitt. However it is taken for granted there that meaningful reference can be made to experiments having yielded one result rather than another. SO instrument readings, or the numbers on computer output, and things like that,

16

are the classical variables of the theory. We have argued already against the appearance of such vague quantities at a fundamental level ..... It was for this reason that the hypothesis was made of fundamental variables x, from which instrument readings and so on can be construct ed.... I suspect Everett and DeWitt wrote as if instrument readings were fundamental only in order to be intelligible to specialists in quantum measurement theory. (Bell 198 1).

This is closely related to another problem: that to specify a probability distribution

over "worlds" we must choose a preferred operator, so that the components of the statevector

along its eigenvectors can be taken as the set of "points" on which the measure is defined.

But for Everett, the different "worlds" are not physically defined "configurations" (Bell's

Everett(?)) or "trajectories" (de Broglie-Bohm), but are worlds-for-consciousnesses. I stress

here the fact of consciousness, not the macroscopic nature of instruments. Macroscopic

superpositions exist; we just cannot, in the nature of things, perceive them. Instrument

readings are important because they are observed by consciousnesses. Thus while the

division into worlds is indeed dependent on vague, nonphysical concepts like consciousness,

these concepts are not imported into the basic physics or the mathematical formalism,

because the grouping of Hilbert-space components into distinct worlds is not a phenomenon

at that level (though it could be said to have a basis at that level in the possibility of

superposition inherent in the QM formalism). Therefore, there is no need in the basic

physical formalism to choose a preferred observable whose eigenvectors determine the nature

of the various "worlds."

(C) is the contention that Everett/de Witt do not succeed in deriving the desired

probability measure over final configurations from the QM formalism. "Everett has to attach

weights to the different branches of his multiple universe, and in the same way [as Bohm]

does so in proportion to the norms of the relevant parts of the wave function. Everett and

De Witt seem to regard this choice as inevitable. I am unable to see why, although of

course it is a perfectly reasonable choice with several nice properties." (Bell 1981). This

17

point is very important, since if we can't get the right probabilities straight from the physical

formalism then we will have to add something to it, which will probably require mention of

a particular decomposition of the statevector into worlds and explicit specification of a

measure over those worlds, thus requiring the extra structure of Bohm and Bell's "classical

configuration variables x." The validity of Bell's contention, and of various attempted

derivations of the probability measure by MWI proponents, are considered at length in

sections 7-9 below.

(D) is the question of trajectories and branching. I have already indicated that

branching is not a process represented in the fundamental physics, and Bell is mistaken to

attribute to Everett the view that it is represented there. Bell also mentions "the [lack of]

association of a particular present with a particular past". I have already discussed this as

well: one may suppose such association occurs in the unification of consciousness, rather

than by explicitly putting trajectories in terms of "configuration variables" into the physics.

6.2 Wheeler, Wigner, and d'Espagnat. Aside from Bell's, the most influential criticisms

of the MWI have probably been those of Wheeler, Wigner, and d'Espagnat. The portions

of their criticisms which are relevant to version (A) of the MWI are quite similar. They

object to the existence of branches of the statevector in which "we" have experiences

different from those we actually have. Thus Wheeler:

...g reat difficulties would seem to arise in giving any well defined meaning to the term 'state of memory of the observer'. Thus as Bohr always emphasized, any attempt to "push the analysis of the mechanism of living organisms [Le. the consciousness as ultimate observing device] as far as that of atomic phenomena ... would doubtless kill [the] animal" [and thus wipe out that very consciousness]. (Wheeler 1977. Brackets Wheeler's).

He then cites Wigner and d'Espagnat. Now it is likely that an attempt to view the

consciousness as part of the quantum-system, and, say, to try to reverse a conscious Stern-

18

Gerlach measurement would, even theoretically within the MWI, wipe out that observer's

memory of what went on between the measurement and its reversal (though not necessarily

the observer-- if one assumes enough finesse to reverse a conscious Stern-Gerlach

measurement, one might as well assume enough finesse to keep the observer alive in the

process). But this does not invalidate the theory. And if one arranged to have it done on

oneself, one might well know after reversal, that one had been in a superposition (statevector

reduction had not occurred), because of the memory gap. Whereas if reduction had occurred,

one would be unable to reverse the measurement and might remain conscious of the

measurement result.

Wigner shows some awareness of the psychological basis of the MWI, as he argues:

The epistemology under discussion postulates that I feel to be in only one of the immensely many states in the linear combination of which I actually am. This means, however, that the other states, and hence the total state vector of the world, are meaningless.

One could postulate, in a similar vein, that the world is homogeneous and isotropic and it is only I who sees differences between different locations and different directions--the real state vector is invariant under all Poincare transformations. Such ideas appear to me--perhaps wrongly--to be detached from reality. (Wigner 1973).

There is an equivocation on 'I' in the first sentence, between 'I' as whichever consciousness

I happen to wind up with after measurement, and 'I' as the physical subsystem which gets

thrown into superposition after measurement. There is no ground for calling the other states

meaningless except an extreme positivism in which only measurement results are real.

There are many reasons to reject such a positivism and try to find a realistic interpretation

of physical theory, and the justification for introducing the "meaningless" other branches into

the MWI is that they enable a realistic (and non-hidden-variables) interpretation of QM.

Superpositions are necessary in order to explain interference phenomena; in this regard they

are unlike the hypothetical supposition of homogeneity and isotropy of the real statevector

19

to which Wigner compares them in the last paragraph: the latter doesn't help explain

anything.

D'Espagnat (1976, Chapter 23) deals with two versions of the MWI, his "solution (a)"

and "solution (b)." Solution (a) roughly corresponds to a physical multiplicity view, my (B)

or (C), to which d'Espagnat objects because of the need to choose a preferred observable.

Solution (b) is one in which "the number of instruments and the number of systems is

always conserved." It would therefore appear to correspond to my version (A), but

d'Espagnat does not appear to recognize the need for a psychological interpretation of the

split.

Thus we are led to the conclusion that, if the impressions of the observers are just physical properties of these systems--and as such, are described by the state vectors--then these impressions must be blurred in the kind of experiments studied here. This conclusion is, however, in obvious contradiction with our immediate experience, since under the conditions of these experiments every one of us knows with certainty that he always has the definite impression of seeing, or not seeing, a signal. (d'Espagnat 1976, p. 27 3)

The problem with d'Espagnat's conclusion (and the argument that leads to it) is the

assumption that there is a single observer consciousness associated with a single physical

observer-system and represented in the statevector (the plural in the first sentence refers to

a hypothetical ensemble of identical universes (d'Espagnat likes to couch his quantum theory

in terms of ensembles), not to the branches of a statevector within a universe). We cannot

expect the statevector to be in a "one-to-one correspondence'' with my single impression;

it corresponds to a multiplicity of impressions, which is not the blurred impression of a

single consciousness, but the distinct impressions of the multiplicity of consciousnesses which

my version (A) says must exist after measurement. To expect the objective description of

physical reality to contain the "fact" that I had this particular impression is like expecting

it to explain why my consciousness is that of being Howard Bamum, and not of being

20

Bernard d'Espagnat: in both cases the explanation is roughly that various subsystems of

physical reality with a certain structure exist, some sort of crude psychology tells us that

each subsystem with such a structure has consciousness, and given that I am going to be a

consciousness, I have to be one of them and can't be more than one at once, but physics

has nothing to say about which one3, nor should we expect it to.

6.3. Thus it would appear that versions (A), (C) or (D) of the MWI remain competitors for

the de Broglie-Bohm hidden variables theory (a variant of G ) , as do hidden variable

interpretation (F) and other variants of (G). There might seem little to go on in making the

choice except for metaphysical prejudice (e.g. against unnecessary multiplication of

consciousnesses, or against "hidden" variables). But there is also the fact that version (A)

does not have to choose a preferred set of configuration variables-- indeed, does not have

to add anything to the Schrodinger equation formalism. This may be viewed as an

advantage or not, depending on whether one agrees with Bell (1982) that "in physics the

only observations we must consider are position observations ... It is a great merit of the de Broglie-Bohm picture to force us to consider this fact. If you make axioms, rather than

definitions and theorems, about the "measurement" of anything else, then you commit

redundancy and risk inconsistency." Speculating, one might even relate this to the possibility

(Wigner 1976) that only a few of the mathematically possible Hermitian operators correspond

to physically measurable quantities like position and momentum. If this is so, the de

Broglie-Bohm measurement theory, and perhaps version (C) of the MWI as well, might

explain it, while the standard theory and version (A) of the MWI do not. (Though Bohr's

'hecessity for using classical variables to describe measurement results" could be considered

3 B ~ t physics combined with the "stylized psychology" of the MWI may enable us to say something about the probabilities of being various consciousnesses, when your current one splits due to measurement. See Section 8.

21

a recognition of it). In addition, there is the fact that we have not, as yet, introduced

probabilities into version (A) of the MWI. If they can be derived from this version which

does not explicitly mention them, this might be considered an advantage. If they cannot be

so derived, we must investigate whether they can be coherently introduced by stipulation (as

they are in the physical multiplicity versions of the MWI). The latter question will be

investigated in section 7, the former in section 8.

7. The Status Of Probabilities in Version (A) of the MWI. Let's put aside until the next

section Everett and DeWitt's contention that the statistical interpretation emerges naturally

from the formalism, and consider whether we may simply introduce the square-amplitude

measure as a matter of stipulation. Does it make sense to do so? What does it mean to

assign probabilities to components of the statevector? If one wants to say it has to do with

relative frequencies, fine-- but relative frequencies of what?

Consider several ways probabilities could be included in a physical theory. (1) Under

certain conditions (unambiguously specified in the theory, one hopes) the theory does not

predict determinate consequences, but makes a statistical prediction. Le., the equations of

motion may contain transition probabilities. For example, particles make "random jumps."

(2) Equations of motion are deterministic, but the initial conditions are probabilistic. This

could be described by saying the actual physical world is conceived of as chosen from a

(possibly hypothetical) ensemble of possible worlds, with probabilities given by a certain

measure. This could be conceived of as (2a)a physically real random process occurring at

the "beginning of the world" in which initial conditions are "chosen." Or it could perhaps

be due to (2b)our ignorance of initial conditions. In the latter case one wonders if it is

proper to call the probabilities part of a genuine physical theory. Perhaps it is, if the

22

physics of the situation explains why we can't know the values and why we should have the

subjective probabilities we do. Statistical mechanics is one case in point; Bohm (1953) has

argued that his hidden variable theory is another.

The MWI does not fit neatly into any of the above categories. With the MWI, there

are determinate initial conditions (initial statevector) and a deterministic law of evolution, so

the probabilities must arise from ignorance. But this is not ignorance of an objective

physical fact, rather it is ignorance as to which branch of the statevector we are in or will

go into--ignorance of a "perspectival fact". In Everett's version of the MWI (my version (A))

the branches "split", rather than being distinct and merely becoming differentiated (section

5) , so that it is not a mere matter of finding out which branch one is in, but a matter of

finding out which one of the branches one will end up in after the split. There is a genuine

stochastic event going on; it is, of course, the splitting of one mind into two, as viewed

from the perspective of one of the resulting conscious histories. The probabilities are

probabilities that "my" consciousness will go with one branch versus the other; and their

introduction can be empirically justified by the fact that if we suppose these probabilities

obtained in previous, similar, situations where Schrodinger evolution predicts a split (quantum

superpositions amplified to affect conscious minds), we get relative frequencies very close

to the ones we've actually observed (at least with very high probability, in the limit as the

number of observations gets large, which is all any interpretation of quantum mechanics can

promise). These probabilities may be thought of as "objective" as opposed to

"epistemological" (or as they are sometimes termed, "subjective") probabilities. But they are

not the probabilities of objective physical events, but rather of outcomes of a sort of

subjective, psychological, perspectival event, the splitting of consciousness.

If the splitting

initially similar worlds

23

of consciousness were simply a matter of physically distinct but

becoming differentiated, and the probability meant something like the

relative numbers of various types of these worlds, the physical basis of the probabilities

would be relatively clear. In these interpretations, the worlds in the "comparison set of

possible worlds" differ physically, and the probabilities specify a distribution over these

different physical possibilities. But in version (A), the probabilities may seem ad hoc,

necessary for agreement with experiment but a nebulous, hard-to-interpret element in a theory

which is, after all, attempting to remedy orthodox quantum mechanics' unclear and/or

idealistic view of the world by providing a clear and precise picture of reality.

The difficulty is not per se that the probabilities are probabilities of perspectival

events rather than probabilities describing events occurring in the world as seen from no

particular point of view. It is that we need a coherent picture of how these probabilities of

perspectival events (e.g. the probability that "my" consciousness will be the one associated

with the "pointer-indicating-z-spin up" component of the measuring apparatus) are grounded

in objective physical reality. After all, for other "perspectival facts," one can generally come

up with a reasonable explanation of why, given the objective physical situation, things

viewed from here had to look as they did. One can of course derive the probabilities from

a coherent mathematical structure. But we need to justify the identification of these numbers

(norms of Hilbert-space components) with probabilities of perspectival events. Since the

basic formalism is supposed to be nonprobabilistic, this interpretation of Hilbert-space norms

as probabilities really should arise from the basic structure and the nature of the perspectives

we can have on it (the notion of splitting consciousnesses), rather than simply being

stipulated. Thus, Everett, de Witt, and Graham's claims to have derived the probabilities

from the Schrodinger equation formalism become crucial.

24

8. Deriving the Probabilities From the MWI.

8.1 Everett's derivation. In deriving the square amplitude measure from the formalism

Everett's main assumption (1957a,b) is the additivity assumption that if ty = Caityi is the component of the wavefunction along the vector w, its measure rn(ty) should equal the sum of the measures associated with each of the components in its expansion, X,m(qvi). This

additivity requirement is justified by a "conservation of probability" argument: we want the

probability of being in the "trunk" of a particular tree to equal the sum of the probabilities

of winding up on each of its branches. But this argument appeals to trajectories, which are

not part of the physical formalism of version (A). One might object therefore that it should

be applied to versions of the MWI of type (E) rather than type (A). To this I respond that

the "trajectories" are to be interpreted as the psychological trajectories discussed in section

4 above. But what grounds do we have to require the additivity property of these trajectories? Well, we may ask if the apparatus that tells us which branches there are

(statevector, Schrodinger equation, and psychological conditions for unity of consciousness)

tells us anything about these probabilities. Let a "memory-sequence" a = &a2, ..., an) be a sequence of observed outcomes of n experiments Note that the sum of probabilities, given

that I am in the branch corresponding to a memory-sequence a, of going into branches where

my memory-sequence is a extended by one of the an+,,k where k indexes the possible

outcomes of the (n+l)th experiment, which is about to be performed, is one. This is a fact

about the way consciousness unifies through time: physical subsystems (in a nontechnical

sense) in which one was about to perform a certain experiment unify with later subsystems

in which that experiment is complete with a certain outcome, to create a consciousness with

a history (or in this case, a consciousness branching into several consciousnesses with

25

histories). If your consciousness starts out in one such system which then becomes

correlated with an amplified quantum phenomenon, you can be sure its history will be

continued on one of the resulting branches. The reasoning is essentially the same as that

which justifies assuming that the continued existence of a physical body and brain with the

right relations to ones body and brain at earlier times ensures ones continued conscious

existence. And the conclusion is just Everetts additivity assumption.

But Everetts derivation of a measure also makes the seemingly ad hoc assumption

that the measure rn of a given component must be a function of its norm ai. (He argues

that it must be a function of its modulus in fact, since we may simultaneously multiply all

weights by an arbitrary modulus-one complex phase factor without changing the physical

system represented). But, we might ask, why not let m be a function from Rk to R, instead of R to R, a function of all the weights in a decomposition, or perhaps even a function of both weights and eigenvectors?

8.2 An improvement on Everetts derivation. To deal with this objection we may keep

the additivity assumption and replace the remainder of Everetts derivation with an appeal

to Gleasons theorem (Gleason 1957). This theorem states that for each additive measure

on the closed subspaces of a Hilbert space H, there is a density operator of the usual

quantum-mechanical sort, from which the probability measure can be derived in the usual

way). Gleasons theorem does not in itself show that we must assign to a system described

by a statevector w the probabilities described by the corresponding density operator I w)(w I (Le. the square amplitudes of the expansion of that statevector in terms eigenvectors of the

measured operator). So that if we had a one-to-one mapping R from statevectors to density

operators, we might generate a different probability measure, by assigning to a system with

statevector w the probabilities corresponding to some density operator other than I w)(w I ,

26

like Q(v). But a simple argument shows that in the MWI, we cannot do this. It is only reasonable that if a statevector has a zero projection on the subspace spanned by a certain

eigenvector, there can be no world corresponding to that subspace in the MWI. (To continue

our analogy between spatial-region and Hilbert-space subspace localization of consciousness,

if a certain spacetime region is empty, we can hardly assign consciousness to its contents.

It seems similarly unreasonable to assign consciousnesses to "empty" subspaces of the Hilbert

space.) Therefore we must assign probability zero to worlds corresponding to such empty

subspaces. It follows from the axioms of probability that we must assign probability one to

a subspace if the statevector lies in that subspace. The only measure satisfying these

conditions and additivity is the square-amplitude measure corresponding to the statevector.

(Consider the one-to-one function from H into the space of density operators mentioned

above. Since density operators are Hermitian, they have real eigenvalues and any density

operator p can be represented as p = C l p , I ~ l ) ( ~ l I , where the vl are an orthonormal basis for H and the p, are between 0 and 1 and sum to 1. Letting the actual statevector be called

v0, and letting the Ivk) be normalized states orthogonal to Iv0) and forming an orthonormal basis together with it, we consider measurement of an observable A which has this basis as

its Set of eigenvectors. Letting oc, be the eigenvalue corresponding to an eigenvector Ivk), we require that PA(& I p), the probability that a measurement of A yields a, if the density matrix is p (calculated according to the usual rules for deriving probabilities from density

matrices) is 0 for k # 0, and 1 for k = 0. This means that C,p,II(~kI~l)112 = 0 for k # 0.

For this to be true, since both the pi and the II(vkIyf,)112 are nonnegative, for each i

either pI or II(yfktvj)l12 must be zero. Since this is true for all i and k, it means that the

only Iv,) which can enter the representation of the density matrix with nonzero weight p1 are those which are orthogonal to all the Ivk) with nonzero k. The only such (normalized)

27

vector is Iyfo), so the only vector which can enter the representation of the density matrix

isIyfo), the actual statevector. The only allowable density matrix is thus lyfo)(yfoI, the

density matrix corresponding to the statevector.) We now have an acceptable alternative

derivation of probabilities from the formalism (stressing the psychological nature of the

trajectories behind Everett's additivity requirement, and adducing Gleason's theorem and the

0-1 probability constraints imposed by the MWI).

8.3 The De-Witt Graham derivation. Even later proponents of the MWI, de Witt and

Graham in particular, felt Everett's arguments for the square-amplitude probabilities to be

inadequate, and presented arguments they felt were better. For instance, there is an argument

attributed by De Witt (1971) to Graham (and well presented in d'Espagnat (1976)). (Everett

states the result, while laying less stress on it than do DeWitt (1971) and Graham (1973)).

In this argument, one examines relative frequencies of results in ensembles of identical

experiments, and shows that in the limit as the number of experiments goes to infinity, the

difference between the observed frequencies and those predicted by standard QM can be

made as small as desired, except for a "negligible" set of cases which approaches zero

measure. The difficulty is that the argument that this set is negligible rests on the fact that

the component of the statevector corresponding to such cases approaches zero norm, and the

appropriateness of the norm as a measure of relative probability is precisely what we are

trying to establish, D'Espagnat seems not to view this as a problem, describing the extra

assumption "that vectors with zero norm correspond to nonexisting branches" as "very

natural." This is indeed a natural assumption, but it does not necessarily imply that

statevectors with very small norm have very small probability, unless one makes the

additional assumption that the relative probability of a branch is a uniformly continuous

function of its relative norm, This sounds pretty reasonable, but no argument is given for

it. A variant of this argument defines a relative frequency operator on infinite tensor

products of Hilbert spaces, i.e on vectors representing possible outcomes of infinite sequences

of measurements (and superpositions thereof). Its eigenvalues are shown to be equal to the

quantum mechanical probabilities. When the formalism is interpreted, this turns out to be

the same argument as that just given, and the same objection arises when one asks why the

mathematical idealization of a relative frequency operator on infinite sequences should be

relevant to actual measurements on finite sequences.

Graham's objection to Everett is worth quoting at length:

Everett gives no connection between his measure and the actual operations involved in determining a relative frequency, no way in which the value of his measure can actually influence the reading of, say, a particle counter. Furthermore, it is extremely difficult to see what significance such a measure can have when its implications are completely contradicted by a simple count of the worlds involved, worlds that Everett's own work assures us must all be on the same footing.

(To be sure Everett argues that the measure ... is unique. But remember that Gleason has shown that the probabilities defined by the Born interpretation, considered as a measure on a Hilbert space, are themselves unique. Nevertheless this (hopefully) does not deter anyone from inquiring into the connection between those probabilities and experiments that measure relative frequency.) (Graham 1973).

Graham then makes roughly the point I made in the previous paragraph about the

DeWitt-Graham-Everett demonstration that relative frequencies converge on the quantum

mechanical probabilities: that it relies on the probability measure it is intended to derive.

Now Graham's objection that "those probabilities" (in Gleason's theorem) or "[Everett's]

measure" are not immediately connected with "experiments that measure relative frequency"

requires some subtlety in its interpretation. If the claim is that the square-amplitude

probabilities need to be related to relative frequencies, then the DeWitt-Graham-Everett

argument does as well as can be expected. Any attempt to require an interpretation of

probabilities in terms of limits of relative frequencies will run up against the problem that

29

these limits turn out to be plims-- probability limits-- so we are stuck with probabilities as

primitives. That is no reason to reject a particular use of probability theory such as

Everett's.

An alternative interpretation of Graham's objection would be that it is not so much

the transition from probability to relative frequency as the transition from measure to

probability that is questionable. The formalism allows one to define this square-amplitude

measure, but who says it tells us anything about probabilities of events? Maybe these events

just happen, and there are no probabilities. Although we could assume these particular

probabilities (and, given the above argument for the uniqueness of the square-amplitude

measure, no others) we are not forced to assume any probabilities, so their introduction is

"going beyond the formalism," as surely as in versions (C) or (D) or de Broglie-Bohm. AS

DeWitt (1971) says: "although reality as a whole is completely deterministic, our own little

corner of it suffers from indeterminism. The interpretation of the quantum mechanical

formalism (and hence the proof of Everett's metatheorem) is complete only when we show

that this indeterminism is nevertheless limited by rigorous statistical laws." By examining

the way our own little corner fits into the whole, we can apparently derive what those

statistical laws would have to be, but it is not clear we can derive that such laws must

apply, unless we are willing to argue that we must assign probabilities when there is a

unique acceptable distribution .4

4Graham (1973) has attempted to derive the quantum mechanical probabilities for experiments measuring macroscopic observables by using the quantum statistical-mechanical assumption of equal a priori probabilities and random phases for eigenstates. If we are willing to accept that consciousness is necessarily a macroscopic phenomenon, then this argument, if valid, would allow us to base the probabilities on foundations no less secure than those of statistical mechanics. However, the major justification for the Gibbs principle is that the statistical mechanics constructed from it works, empirically. Perhaps the fact that the quantum mechanical probabilities derived from the mere assumption that we must have some probabilities also work empirically, makes that assumption equally secure, so that there

30

9. The Option of Refraining From Assigning Probabilities. The option of claiming that

the MWI formalism still gives us no picture of how these probabilities of perspectival events

arise, and perhaps even no grounds for assuming such probabilities exist, could correspond

to "complete ignorance"or "true uncertainty" in the decision-theoretic sense (Luce and Raiffa

(1957, pp. 12, 275-8)), or as I will call it, "nonprobabilistic indeterminacy".

There are several possible arguments against this.

(1) Assuming there are no probabilities makes the theory empirically inapplicable,

while the mere assumption that there are some probabilities, plus the Schriidinger

equation/statevector formalism, yields the empirically correct probabilities. If we take this

approach, we fail to explain what it is about the underlying reality described by the MWI

that makes us choose the probabilistic, rather than the nonprobabilistic, version of the MWI.

Furthermore, nonprobabilistic indeterminacy may not be very useful for making predictions,

but as I shall argue below, it is not empirically inapplicable.

(2) Nonprobabilistic indeterminacy is necessarily "subjective" in the sense of

"subjective (i.e. epistemological) probability." It therefore does not apply here, where the

probabilities, as argued in section 7, are "objective" probabilities (though of "subjective", i.e.

perspectival, events). This is to claim that nonprobabilistic indeterminacy is always a matter

of having no clue about the real probability distribution (possibly degenerate), which always

exists. The only justification for not assigning a single distribution would be suspension of

judgement between a number of possible distributions. But the uncertainty in the many-

worlds account of measurement is not due to our ignorance of something now determined.

There is no "fact of the matter" before a measurement, about which branch I will go into,

any more than there is a "fact of the matter," (except perhaps in some abstruse metaphysical

is no need to appeal to statistical mechanics and require consciousness to be macroscopic.

31

sense) what the outcome of a measurement will be on a statevector-reduction view. The

probabilities (or the uncertainty, if we refuse to assign probabilities) are as good as our

knowledge could possibly be, theoretically, and in that sense "objective". (See Section 7.)

A variant of this argument points out that these events are repeatable, and thus not the kind

of "unique event" which it is sometimes argued is the proper context for nonprobabilistic

indeterminacy.

I believe this argument is simply wrong. One might cite, for example, the existence

of infinite sequences with no limits of relative frequencies to buttress the possibility that

"objective" nonprobabilistic indeterminacy could really exist. To counter this, one could

argue that such sequences may result (though with probability zero-- which does not mean

impossibility) from probabilistic processes. And one might argue that no finite amount of

data can convince us we face such an objective nonprobabilistic indeterministic process,

rather than a probabilistic one. Nonprobabilistic indeterminacy would then be acceptable

only as a description of our ignorance in situations where we must act without adequate

data-- never as part of a description of the world which is a candidate for a scientific theory,

viewed as potentially standing up to indefinite amounts of further data collection. Since the

MWI is viewed as a candidate for a (possibly final) scientific theory, and since the

measurements situations which create (perspectival) indeterminacy within it are repeatable,

we would have to view these measurements as situations of probabilistic indeterminacy.

The problem with this is that we can in theory have perfectly good evidence that we

are facing a nonprobabilistic indeterminate process, even on the basis of a finite amount of

data. With an infinite amount of data, we might of course have a sequence with no limit

of relative frequency, and since this has zero probability on any probabilistic hypothesis, we

would be justified in assuming a nonprobabilistic process-- which has nothing to say about

32

the probability of the data, of course, but at least doesn't assign it probability zero. But

there are also finite data sets which justify the inference that it is extremely unlikely they

were produced by a probabilistic process. For instance, divide the data into two sets of

roughly equal size by some previously chosen, data-independent rule. If (for large amounts

of data N) the relative frequencies in the two subsets differ too greatly, we may conclude

the data was probably generated by a nonprobabilistic process. For any probabilistic process

has a very small probability for this sort of discrepancy-- indeed the probability approaches

zero for large N.

(3) We can empirically determine relative frequencies, and perhaps this fact is in

itself sufficient to justify assuming that there are probabilities. This is not adequate because

the task of the MWI is precisely to find a theoretical structure which reproduces the

empirical facts, and this includes reproducing the empirical fact that there are probabilities.

We want a theoretical argument for this from within the MWI.

Conclusion. By taking the "multiplicity of worlds" to occur at the level of consciousness,

Le. of a multiplicity of perspectives on a single reality, I have shown that the MWI is a

coherent and consistent interpretation of quantum mechanics. Many of the arguments that

the MWI is inconsistent were shown to result from the erroneous notion that additional

physical structure was needed to account for the multiplicity of worlds. The notion, put

forth by Bell, for instance, that the quantum mechanical probabilities were only one of many

probability distributions which could be imposed on the MWI was also shown to be incorrect

when applied to the "psychological" version. But an important, possibly fatal, difficulty was

shown to remain: that of vindicating this unique acceptable probability distribution by

showing that so far, no adequate argument is apparent which uses the MWI to rule out

33

nonprobabilistic indeterminacy. Those who take a dubious view of the notion of

nonprobabilistic indeterminacy (some Bayesians, for instance) may see no difficulty here.

I view it as a major difficulty which should be a focus of future work on the MWI.

Even if it is consistent and succeeds in reproducing the experimental facts, there are

reasons for disliking the MWI, for instance that it implies the view that our minds

occasionally split into many minds, which we will probably never be able to reestablish

contact with, in different branches of the "universal wavefunction." While I certainly don't

consider it a priori impossible that science should lead us to this belief--as proponents of the

MWI believe it does--1 do consider this belief sufficiently bizarre a priori that it is worth

investigating the alternatives. However, it may be more scientifically satisfying to adopt the

splitting-consciousnesses view along with a physical theory that is rigorous, precise, and

economical, than to adopt the "reduction of the statevector" view which is uneconomical in

that it postulates two fundamental and radically different modes of change of the statevector,

and fuzzy in that it gives no clear physical criteria for when one mode rather than the other

obtains. Therefore interpretations which also reject statevector reduction appear the most

promising alternatives to the MWI. I argued that the de Broglie-Bohm-Bell type of hidden

variable theory, Bell's Everett(?) version of the MWI (C), its hidden-variables analog (F),

and perhaps some many-worlds-with trajectories (D) interpretations remain logically

acceptable alternatives. Their major differences with version (A) are the need to introduce

probabilities explicitly as part of the physics, the need to add something to the Schrodinger

formalism, and the related need to choose a preferred observable. Once it is necessary to

do this, one might argue, with Bell, that there is little point in retaining the extra worlds;

if so, the major alternatives are version (A) of the MWI and the de Broglie-Bohm-Bell pilot

34

wave t h e ~ r y . ~ Future work needs to examine the consequences and justification of the de

Broglie-Bohm-Bell introduction of preferred observables.

Despite my defense of the consistency and coherence of the MWI, my own

sympathies are more with de Broglie-Bohm-Bell: I find the multiplication of consciousnesses

in the MWI unappealing. This might be termed a mere metaphysical preference. But it is

possible that the ultimate best choice of interpretation will be determined by its fruitfulness

as the basis for new physical theories connecting quantum theory with gravitation, and

"metaphysical" preferences between interpretations in part express different views on their

fruitfulness in this endeavor. It is perhaps more likely, as Bell often suggests, that some

wholly new way of viewing quantum theory will emerge out of such a synthesis. The

currently available realistic interpretations of quantum theory may then be viewed as giving

us suggestions about the nature of this future perspective on quantum theory, since they give

us different pictures of reality which are consistent with current quantum theory and the

experimental facts. The psychological version of the MWI suggests the most conservative

possibility: that the basic structure of quantum theory may survive intact and without major

additions-- at the price of a multiplication of consciousnesses. Hidden variable theories like

de Broglie-Bohm suggest that a more radical revision of the quantum formalism, adding

additional structure beyond a wavefunction framework, may eventuate.6 But whatever one's

preferences, if one wants an accurate idea of the various kinds of realities that are

compatible with quantum theory, it is important to give the psychological version of the

'(F) is vulnerable to Bell's accusations of solipsism of the present moment and seems to have no compensating advantages over any of the other versions. I omit (E) on the grounds that preferred observables must be chosen on arbitrary, "human-centered'' grounds. See Bell (1981).

The two types of theory also differ with respect to the nature of "nonlocality" or "nonseparability" they involve, but there is not space to treat this here.

35

MWI a place among consistent interpretations of that theory, and to understand the true

nature of the remaining difficulties with it.

36

REFERENCES

Aharonov, Y., and Albert, D. (1980) "States and Observables in Relativistic Quantum Field

Theories", Physical Review D 21: 3316-3324.

Aharonov, Y., and Albert, D. (1981), "Can We Make Sense out of the Measurement Process

in Relativistic Quantum Mechanics?", Physical Review D 24: 359-370.

Bell, J.S. (1975), "On Wave Packet Reduction in the Coleman-Hepp Model", Helvetica Physica

Acta 48: 93-8. Reprinted in Bell (1987).

Bell, J.S. (1976), "The Measurement Theory of Everett and de Broglie's Pilot Wave", in

Quantum Mechanics, Determinism, Causality, and Particles, M. Flato et al. (eds.).

Dordrecht, Holland: D. Reidel, pp. 11-17. Reprinted in Bell (1987).

Bell, J.S. (1981), "Quantum Mechanics for Cosmologists", in C. Isham, R. Penrose, and D.

Sciama (eds.), Quantum Gravity 2. Oxford: Clarendon Press, pp. 61 1-37. Reprinted

in Bell (1987).

Bell, J.S. (1982), "On the Impossible Pilot Wave", Foundations of Physics 12: 989-99.

Reprinted in Bell (1987).

Bell, J.S. (1987), Speakable and Unspeakable in Quantum Mechanics. Cambridge University

Press, 1987.

Bell, J.S., and Nauenberg, M. (1966), "The Moral Aspect of Quantum Mechanics", in A. De

Shalit, H. Feschbach, and L. Van Hove (eds.), Preludes in Theoretical Physics.

Amsterdam: North Holland, pp. 279-86. Reprinted in Bell (1987).

Bohm, D. (195 l), Quantum Theory, Chapter 22. Englewood Cliffs NJ: Prentice-Hall. (Dover

reprint, Mineola NY, 1989).

37

Bohm, D. (1952), "A Suggested Interpretation of Quantum Mechanics in Terms of 'Hidden

Variables'", Parts I and 11, Physical Review 85: 160-93. Reprinted in Wheeler and

Zurek (1983).

Bohm, D. (1953), "Proof that Probability Density Approaches tf in Causal Interpretation

of Quantum Theory", Physical Review 89: 458-466.

Daneri, A., Loinger, D., and Prosperi, G.M. (1962). "Quantum Theory of Measurement and

Ergodicity Conditions", Nuclear Physics 33: 297-3 19. Reprinted in Wheeler and Zurek

(1983).

Davies, P.C.W., and Brown, J.R. (eds.) (1986) The Ghost in the Atom. Cambridge: Cambridge

University Press.

d'Espagnat, B. (1976), The Conceptual Foundations of Quantum Mechanics. 2nd ed. Reading,

MA: Addison-Wesley. (Addison-Wesley reprint 1989).

De Witt, B.S., and Graham, N. (eds.), The Many-Worlds Interpretation of Quantum Mechanics.

Princeton University Press, 1973.

De Witt, B.S. (1971)., "The Many-Universes Interpretation of Quantum Mechanics", in B.

d' Espagnat (ed.) Proceedings of the International School of Physics ' Enrico Fermi'

Course IL: Foundations of Quantum Mechanics. NY: Academic Press. Reprinted

in De Witt and Graham (1973) and Wheeler and Zurek (1983).

Everett, H.D. (1957a), "'Relative State' Formulation of Quantum Mechanics", Reviews of

Modern Physics 29: 454-62. Reprinted in De Witt and Graham (1973) and Wheeler

and Zurek (1983).

Everett, H.D. (1957b), The Theory of the Universal Wave Function, Ph.D. thesis, Princeton.

Reprinted in De Witt and Graham (1973).

38

Englert, B.-G., Schwinger, J., and Scully, M.O. (1988). "Is Spin Coherence Like Humpty-

Dumpty? I", Foundations of Physics 18: 1045.

Fine, A. (1973), British Journal of the Philosophy of Science 24: 33-34.

Gleason, A.M. (1957), "Measures on the Closed Subspaces of a Hilbert Space", Journal of

Mathematics and Mechanics 6: 885-893.

Graham, N. (1973), "The Measurement of Relative Frequency", in De Witt and Graham

(1973), pp. 229-253.

Luce, R.D. and Raiffa, H. (1957), Games and Decisions. New York: Wiley. (Dover reprint

1989).

Peres, A. (1980), "Can We Undo Quantum Measurements", Physical Review D 22: 879-83.

Reprinted in Wheeler and Zurek (1983).

Scully, M.O., Englert, B.-G., and Schwinger, J. (1989), "Is Spin Coherence Like Humpty-

Dumpty? 111. Effects of Observation and Correlation", preprint.

Schwinger, J., Scully, M.O., and Englert, B.-G. (1988), "Is Spin Coherence Like Humpty-

Dumpty? 11", Zeitschrift fur Physik D IO: 135.

Wheeler, J. (1977), "Include the Observer in the Wave Function?", in J. k i te Lopes and M.

Paty (eds.), Quantum Mechanics a Half Century Later, Dordrecht, Holland: D. Reidel,

pp. 1-18.

Wigner, E.P. (1961), "Remarks on the Mind-Body Problem", in I.J. Good (ed.), The Scientist

Speculates, New York: Basic Books. Reprinted in E. Wigner, (1967), Symmetries and

Reflections, Bloomington: Indiana University Press, and in Wheeler and Zurek (1983).

Wigner, E.P. (1963), "The Problem of Measurement", American Journal of Physics 31: 6-

15. Reprinted in E. Wigner (1967), Symmetries and Reflections. Bloomington:

Indiana University Press, and in Wheeler and Zurek (1983).

39

Wigner, E.P. (1973), "Epistemological Perspective on Quantum Theory", in C.A. Hooker (ed.),

Contemporary Research in the Foundations and Philosophy of Quantum Theory.

Dordrecht, Holland: D. Reidel, pp. 369-385.

Wigner, E.P. (1976), "Interpretation of Quantum Mechanics", mimeographed notes, revised

and reprinted in Wheeler and Zurek (1983), pp. 260-314.

Zeh, H.D. (1970), "On the Interpretation of Measurement in Quantum Theory", Foundations

of Physics I: 69-76. Reprinted in Wheeler and Zurek (1983).